CASIA OpenIR  > 毕业生  > 硕士学位论文
面向Massive MIMO的波束赋形算法研究及实现
王戈
学位类型工程硕士
导师尹志刚
2017-05-25
学位授予单位中国科学院研究生院
学位授予地点北京
关键词Massive Mimo 天线子阵列 数字波束赋形 混合波束赋形 专用代数指令提取
摘要Massive MIMO关键技术由于能够提供更高的信道自由度,通过空间域的扩展提高系统的频谱及能量效率,成为5G关键技术研究重点。但是,Massive MIMO关键技术通常在基站侧需要部署数十根乃至上百根天线,导致基站侧发送信号在波束赋形处理过程中的运算复杂度大幅度增加,因此极大地制约了Massive MIMO关键技术在5G通信系统中的实际应用。同时,Massive MIMO关键技术还面临由于天线个数的增加,而造成的高成本问题。在这样的背景之下,本论文通过设计低复杂度、低成本的波束赋形算法用以解决Massive MIMO关键技术的应用瓶颈。除此外,本论文根据所设计的Massive MIMO波束赋形算法,提取波束赋形算法面向UCP处理器的专用代数指令,从而进一步从波束赋形算法实现角度攻克Massive MIMO的工程应用壁垒。
本论文通过引入天线子阵列的Massive MIMO通信系统结构,重新构造波束赋形权值矩阵,将此前复杂的求取波束赋形矩阵以获取最大接收功率的过程,转化为求取相独立的不同天线子阵列的最优波束赋形列向量。此外,本论文采用幂代法,代替此前复杂度高且难并行化的SVD算法。复杂度分析及仿真结果表明,本论文提出的数字波束赋形算法,相较于传统的SVD算法而言,运算量降低至少一半以上。同时,随着迭代次数增加,本论文提出的波束赋形算法在获取更高增益的同时不会大幅增加运算复杂度。
为了进一步解决由于数字波束赋形算法带来的高成本问题,本论文在提出的数字波束赋形研究基础上引入模拟域的波束赋形。本论文通过数字域波束赋形,微调模拟域的波束赋形结果,从而最小化混合波束赋形算法与论文提出的数字波束赋形间的均方误差。复杂度分析及仿真结果表明,论文提出的混合波束赋形算法在降低硬件成本的同时,具有低复杂度的优势。
同时,为了提高处理器效率,减小信号时延,本论文提取波束赋形算法主要算子,基于UCP处理器特征及其指令集进行算法解决方案的设计,经流水线的优化后提取代数指令。微码运行周期统计结果表明,基于UCP处理器的波束赋形专用代数指令能够支持5G毫秒级端到端的时延通信要求。
其他摘要Due to the Massive MIMO technology is able to provide higher channel freedom and improve the spectrum efficiency, it has been paid more and more attentions. However, the Massive MIMO teconology, requiring dozens or even hundreds of antennas at the base station, produces significant increase in computational complexity in the signal process of beamforming procedure, which has greatly restricted the Massive MIMO development in the practical application of 5G systems. In addition, the Massive MIMO technology is confronted with the increase in the number of antennas, resulting in high cost preoblems. Under these backfrounds, the paper aims to solve the bottleneck of Massive MIMO technology by designing low-complexity and low-cost beamforming algorithm. In addition, this paper designs the parallelization based on the UCP processor characteristics and instruction sets to overcome the engineering application barrier of Massive MIMO teconology.
According to the system based on the sub-array structure, this paper, by reconstructing the beamforming matrix, decomposes the process of obtaining the beamforming matrix to get the maxiumum received power into a series of ones much easier to be solved by optimizing each antenna array one by one. In order to get the beamforming vector of each sub-array with fewer computational complexity, this paper proposes to utilize the power iteration algorithm to replace the high-complexity SVD altorithm with difficulty of parallelism. According to the detailed complexity analisis and simulation results, compared with the traditional SVD beamforming algorithm, the complexity of the proposed digital beamforming algorithm based sub-array structure is reduced by at least half. Meanwhile, with the increase of the number of iterations, the digital beamforming algorithm proposed in this paper could obtain necessary performance gain without significantly increasing the computational complexity.
In order to solve the high-cost problem caused by the digital beamforming algorithm, this paper introduces the analog beamforming on the basis of the proposed digital beamforming. In this paper, the digital beamforming adjusts the results made by analog beamforming to minimize the mean square error between the hybrid beamforming algorithm and the proposed digital beamforming. It is showed the proposed hybrid beamforming algorithm, which has the ability to obtain the near-optimal system performance, can reduce the hardware cost, and inherit the low-complexity advantage of the proposed digital beamforming algorithm.
In order to optimize the pipeline efficiency and reduce the signal process delay, this paper extracts the main operator feature of these two altorithm, and designs the parallelization based on the UCP processor characteristics and instruction sets. Then this paper gives several pratical solutions to extract algebra instructions. According to the algrbra instructions cycles of different beamforming units, it is shown that the extracted beamforming algebra instructions based on UCP processor can support the requirements of 5G system to achieve the millisecond end-to end delay.
文献类型学位论文
条目标识符http://ir.ia.ac.cn/handle/173211/14636
专题毕业生_硕士学位论文
作者单位中国科学院自动化研究所
推荐引用方式
GB/T 7714
王戈. 面向Massive MIMO的波束赋形算法研究及实现[D]. 北京. 中国科学院研究生院,2017.
条目包含的文件
文件名称/大小 文献类型 版本类型 开放类型 使用许可
答辩版本.pdf(2740KB)学位论文 暂不开放CC BY-NC-SA请求全文
个性服务
推荐该条目
保存到收藏夹
查看访问统计
导出为Endnote文件
谷歌学术
谷歌学术中相似的文章
[王戈]的文章
百度学术
百度学术中相似的文章
[王戈]的文章
必应学术
必应学术中相似的文章
[王戈]的文章
相关权益政策
暂无数据
收藏/分享
所有评论 (0)
暂无评论
 

除非特别说明,本系统中所有内容都受版权保护,并保留所有权利。